#asahi-gpu · 2021-05-25 02:54
bloomwell#asahi-gpu · 2021-05-25 02:54
bloomdougall: it's everything :)#asahi-gpu · 2021-05-25 02:52
bloomdougall: ^ dEQP-GLES2.functional.shaders.loops.*#asahi-gpu · 2021-05-25 02:51
bloom Passed: 625/625 (100.0%)#asahi-gpu · 2021-05-25 02:29
bloomdougall: You did the hard part I just typed it in :p#asahi-gpu · 2021-05-25 01:19
bloommore work remains but hey#asahi-gpu · 2021-05-25 01:17
bloomthanks for the delicious notes#asahi-gpu · 2021-05-25 01:17
bloomdougall: Passed my first test with loops#asahi-gpu · 2021-05-24 22:09
bloomwrong window#asahi-gpu · 2021-05-24 18:29
bloom!ss#asahi-gpu · 2021-05-24 12:49
bloomwith do...while as the loop priitives#asahi-gpu · 2021-05-24 12:49
bloomAGX is really, really weird in that its GPU asm is based on structured CF#asahi-gpu · 2021-05-24 12:49
bloomyrlf: nod#asahi-gpu · 2021-05-24 02:49
bloomheh#asahi-gpu · 2021-05-24 02:31
bloomstudies dougall 's notes#asahi-gpu · 2021-05-24 01:48
bloomfor at least 2 iterations you come out ahead with the Apple idiom#asahi-gpu · 2021-05-24 01:48
bloomactually the dummy while_fcmp isn't needed but the point still stands#asahi-gpu · 2021-05-23 23:50
bloomso for a single iteration they're the same, for 2 or more the Metal one is faster#asahi-gpu · 2021-05-23 23:50
bloomwhereas the Metal way, despite adding 2 instructions _outside_ the loop, has a body only of "..., while_fcmp, jmp_exec_any"#asahi-gpu · 2021-05-23 23:50
bloomthe literal NIR way has a body of "if_fcmp, break, pop_exec, ..., while_fcmp, jmp_exec_any"#asahi-gpu · 2021-05-23 23:49
bloomwhy? the loop body is much smaller#asahi-gpu · 2021-05-23 23:49
bloomthe number of instructions is the same between the two -- HOWEVER, the Metal translation should be faster#asahi-gpu · 2021-05-23 23:48
bloomwhich is the natural translation from the NIR `loop { if (!x) { break } .. }`#asahi-gpu · 2021-05-23 23:48
bloomof course, it could also become "do { if (!x) { break } ... } while(true)"#asahi-gpu · 2021-05-23 23:44
bloomwhich means "while(x) { .. }" becomes "if (x) { do { ... } while (x) }"#asahi-gpu · 2021-05-23 23:43
bloomI guess the primitive is "do { ... } while(x)" as opposed to plain while#asahi-gpu · 2021-05-23 23:29
bloomI admit I'm confused by the idioms the Metal compiler uses for loops#asahi-gpu · 2021-05-23 15:50
bloom:D#asahi-gpu · 2021-05-23 15:46
bloomfinish your hypervisor, start the DRM driver, see what I care ;-p#asahi-gpu · 2021-05-23 15:46
bloomI'm kinda over it#asahi-gpu · 2021-05-23 15:46
bloomand tomorrow might be a stat holiday so I might even get something done#asahi-gpu · 2021-05-23 15:45
bloomwell, no excuse not to AGX today then!#asahi-gpu · 2021-05-23 15:45
bloommarcan: thank you uwu :p#asahi-gpu · 2021-05-23 15:39
bloommarcan: can i haz op? :-p#asahi-gpu · 2021-05-23 15:38
bloomHello hello :)#asahi-gpu · 2021-05-12 19:53
bloomAh#asahi-gpu · 2021-05-12 17:57
bloom is macOS dmesg distinct from Console.app?#asahi-gpu · 2021-05-12 16:05
bloommacOS has dmesg?#asahi-gpu · 2021-05-10 19:04
bloomAnd if something goes horribly wrong I can file a CVE, win-win ;-p#asahi-gpu · 2021-05-10 19:04
bloomsegher: I figure poking randomly from unprivileged userspace in a machine with SIP/etc in tact is fair game#asahi-gpu · 2021-05-10 13:15
bloomcompute shaders will happen wen they happen, easier tha 3D but less fun :-p#asahi-gpu · 2021-05-10 03:12
bloomLet's play spot the bug 🤦♀️#asahi-gpu · 2021-05-10 03:12
bloom...#asahi-gpu · 2021-05-10 03:12
bloom struct hash_entry *he = _mesa_hash_table_search(so->variants, &key);#asahi-gpu · 2021-05-10 03:11
bloom....#asahi-gpu · 2021-05-10 03:11
bloomagx_update_shader(struct agx_context *ctx, struct agx_compiled_shader **out, enum pipe_shader_type stage, struct asahi_shader_key *key)#asahi-gpu · 2021-05-08 16:56
bloomAt this pass just interpret as a fail. But flakes make continuous integration a mess.#asahi-gpu · 2021-05-08 16:55
bloomjix: Running the test multiple times in a row and sometimes pass, sometimes fail#asahi-gpu · 2021-05-08 16:55
bloom84.5%, that's an A- at U of T! :-p